Memory-based reservoir computing for synchronization and its applications

Area of Science:

• Complex Systems Dynamics
• Machine Learning
• Information Security

Background:

• Reservoir computing (RC) is a powerful framework for modeling complex dynamical systems.
• Traditional RC models often struggle with systems exhibiting hereditary properties.
• There is a need for advanced RC architectures capable of enhanced prediction and synchronization.

Purpose of the Study:

• To propose a novel memory-based reservoir computing (RC) architecture.
• To enhance the prediction capabilities for chaotic systems.
• To explore synchronization properties and develop secure communication applications.

Main Methods:

• Incorporation of an explicit linear autoregressive memory mechanism into the RC architecture.
• Evaluation of prediction accuracy and valid prediction times on chaotic systems.
• Analysis of synchronization behavior in a drive-response configuration under noisy conditions.
• Development of a chaos-based image encryption scheme leveraging RC synchronization.

Main Results:

• The memory-based RC architecture demonstrated superior prediction accuracy and longer valid prediction times compared to classical RC.
• Heterogeneous memory-based RCs achieved reliable and robust synchronization, even with observational noise.
• The proposed architecture exhibited faster synchronization and smaller steady-state errors than conventional RC.
• The chaos-based image encryption scheme effectively obscured visual content and resisted statistical attacks with lossless decryption.

Conclusions:

• Synchronization is an emergent behavior in trained memory-based RC systems.
• The proposed memory-based RC architecture offers significant improvements in prediction and synchronization.
• This framework provides a promising approach for secure communication applications, particularly in chaos-based image encryption.